But Lewis, What We Really Need is Instant Runoff Voting

Lewis Black on the American political system.

Our two-party system is a bowl of shit looking in the mirror at itself. … Basically, the last eight years, I feel, the Republicans stood around farting and the Democrats went, ‘Ooh, let me smell it.’

30 thoughts on “But Lewis, What We Really Need is Instant Runoff Voting”

  1. I can’t tell if you’re being facetious, but in case you’re not, instant runoff voting will almost certainly result in the same sort of two-party dominated system that we have now with plurality.

    Australia has used it for years, and they only have two parties in their house of representatives. (Well, one party and one 40-year old coalition agreement.)

    If you’re really interested in electoral system reform, approval voting and score voting are a better bet.

    leastevil.blogspot.com <- my blog on the issue
    scorevoting.net <- excellent resources, including comparisons to IRV

  2. Not being facetious at all, Dale. IRV is the only alternative system I’ve actually heard of. I will definitely further research the options you’ve mentioned and may need to take a closer look at the legislature in Australia.

    I think we can agree, though, that as long as the United States continues with its system of winner-take-all voting, third-party candidacies will not be a feasible reality. It will continue to resemble Black’s narcissistic bowl of shit.

  3. Glad to hear it!

    It seems to me that a lot of people are in the same position as you: aching for some kind of change, but only being familiar with IRV. And while the backers of IRV (a group called “FairVote”) have been much more successful in getting their idea publicized, even though the analysis just doesn’t come out in their favor, by virtue of name recognition it manages to stay quite popular.

  4. While it’s true that you can make an equivalent range ballot out of a ranked-order ballot by using a weighting factor, I don’t see how you can go the other way; this is because range ballots are inherently more descriptive.

    And even clone-proof Condorcet methods have incentives for voters to strategically dishonestly rank candidates; the effect is less pronounced than under IRV and plurality, but, since it’s still there, voters will still have to contend with the fear of electing the least-preferred candidate unless they vote for the “lesser of two evils”, which means third-parties will continue to be left out in the cold.

    This, and more, is all better-covered here:
    http://scorevoting.net/rangeVcond.html

    I used to be a ranked pairs fan myself, until I read Gaming the Vote and began investigating Dr. Smith’s work at scorevoting.net, so please, do look in to it! If you get as far as reading his recent paper mathematically proving that no ranked-order ballot based method can possibly be both clone-proof and immune to favorite betrayal, I’m sure you’ll be convinced.

  5. Range ballots are not inherently more descriptive, if the range is set equal to the number of candidates then it is exactly isometric to a ranked ballot. Expanding the range does enable the voter to express some degree of how much he or she prefers a candidate, but requires more strategy. I do not see the added complexity to be worth much, when a ranked ballot is easily understood. I do not think that an approval-style ballot would be as acceptable to voters. Certainly you can find problems with any voting method. IRV fails the monotonicity criterion, but clone-proof Condorcet does not, it is in every respect superior to plurality voting.

  6. I think you’re misunderstanding how a range ballot is used.

    If there are, for example, 5 candidates and 5 values, you don’t have to give one candidate a 5, one a 4, one a 3, etc.; you can give any score to any candidate, including giving the same score to multiple candidates.

    That’s why approval voting is considered a sub-set of range voting; it has two levels, but you can still have more than two candidates.

    So yes, they ARE inherently more descriptive, because you can, with sufficient levels, duplicate any ranked-ballot-plus-weighting-function you could think of, but the range ballot would still allow further discrimination and furthermore allow different voters to use different functions.

    And as for strategy, first you have a lot of terms to define: what’s “more strategy” mean? Because every voting system–even your clone-proof Condorcet–allows voters to strategize to maximum the outcome of election. These precise arguments are why Smith ran the voting simulation that he did and allowed for the full range of from completely-honest voters to maximally-strategic voters. And he found that, with maximal strategy, IRV is precisely as bad as plurality, Condorcet methods are a hardly a hair better, but that range and approval are surprisingly better (it was this result that inspired him to found the Center for Range Voting to push for it’s adoption; note that the simulation came BEFORE he made his decision.)

    I also find it surprising that anyone could claim an approval ballot (“for each of these choices, give a thumbs-up or thumbs-down”) is more complicated than a ranked-order ballot (“arrange these choices from most-favorite to least-favorite”). There’s a reason amihotornot asks for thumbs up-or-down; they found they got more page views (and so, more ad revenue) by asking that, rather than asking people to choose which of two pictures was hotter. And when it comes to score, there’s a reason that IMDB doesn’t ask people to arrange all 100,000+ movies in its database from best to worst, but rather to rate them from 1 to 10. Rating is easier than ranking.

    And sure, CSSD is clone-proof… but it fails both participation and consistency. Sure, every system has its drawbacks. But range voting beats Arrow’s impossibility theorem! Surely that means somthing. http://www.rangevoting.org/ArrowThm.html

    Please, read the links I’ve posted in this reply and my previous ones: all your concerns have been brought up before, and there is hard data supporting my stance.

    It’s beautifully summarized in one image here: http://www.rangevoting.org/BayRegsFig.html

    If you look at nothing else I’ve said, at least look at that figure.

  7. Approval and range voting allow for voters to rank candidates equally, as you pointed out. This is also true of ranked ballots for those candidates a voter does not explicitly choose, so for instance if there were 5 candidates A,B,C,D,E and the voter marked a first choice for B, leaving the rest of the ballot blank, the clear intention of the voter would be to mark B>A=C=D=E. But requiring the voter to strictly order his or her votes may be a good thing, as we are asking the voters to make a decision and exercise their franchise, we are not taking a popularity survey in fact.

  8. Clone-proof Condorcet methods are least subject to strategic voting as long as the count is not disclosed until the close of voting. Voters are most inclined to vote their honest preferences, and this leads further to more honest political advocacy and a saner political culture. CSSD is used in real world organizations, and has been proven. Approval voting is not generally used, and where range voting is most applicable are things like athletic and dancing competitions where the votes are public so less likely to be strategic.

  9. It occurs to me as well, when I lived in Pennsylvania they use approval voting for judges. Retention elections are always won. There is really no way for anyone to contest an incumbent barring gross abuse of office. Approval voting is a terrible thing for democracy.

  10. Equally ranked first choices might as well be considered a spoiled ballot anyhow. The idea that we’re just going to take the highest scoring candidate and let “voters” vote for both or all candidates equally is an absurd mockery of the franchise.

  11. I’m just going to reply to all four of your comments at once, Mike:

    One, I’m not refuting Arrow’s theorem. If you were to examine Arrow’s work, you would notice that his definition of a voting system is that it converts a set of individual’s ranked-order preferences (“votes”) to a single societal ranked-order preference.

    But since approval voting and range voting don’t use ranked-order ballots, Arrow’s theorem cannot be directly applied to them.

    The link I provided earlier http://www.rangevoting.org/ArrowThm.html states this clearly, and then attempts to follow the spirit of Arrow’s work while applying it to non-ranked-order (i.e., cardinal) ballots.

    It also discusses a similar approach to G-S, and even a few other “impossibility” theorems, and provides links to academic papers as sources. It may help to examine the links I provide before attempting to counter-argue them.

    Second, you mention equal rankings in ranked-order ballots; but that equal-ratings are permissible in range (and approval) isn’t what makes them effective. It’s more that each candidate can be evaluated independent of the others and, in the case of range, virtually any degree of strength of preference can be indicated.

    You say “forcing” a voter to indicate a full ordering “may” be helpful; I disagree. I’ll bring up Australia again, where voting is compulsory, and a full-ordering is required for a valid ballot; but a voter can instead “vote above the line”, indicating that they choose an ordering pre-selected by the party bosses of the party they indicate. 85% choose to reject the majority of their franchise and vote above the line: because ranking is cognitively difficult (and the use of IRV means most of that effort would be wasted anyway.) Ratings are, comparatively, cognitively easier; which is WHY they so often used to judge sporting events; you’ll note that most elections only have two viable candidates, while judged sporting events have many more than two.

    And while the intent of a judged event or a popularity contest is different than the intent of an electoral ballot, the underlying computational problem is identical, and so the same algorithms can be applied to both.

    And now: a reiteration of strategy. Are you familiar with polling? Truthfully, we are bombarded with “+-3% margin of error” election results daily in the lead up to important elections, and there is a whole industry devoted to providing them. Not only is this polling data profitable to provide, it is useful to voters, and any attempt to make a Condorcet method the law of the land will not be able to outlaw the practice.

    Honesty is a powerful electoral tonic. Every voting method leads to better results the more honest its voters are. But strategy can be quite powerful for each individual voter, hence the tension. That’s why, again, you should read the links I provide before counter-arguing them, particularly this one: http://www.rangevoting.org/BayRegsFig.html For which I said, “If you look at nothing else I’ve said, at least look at that figure.” It examines honesty and strategy for several selected voting systems and plots the quality of their average outcome. You will note that, yes, Condorcet methods are much better when everyone is honest. But you will also note that, when everyone is honest, approval and range are still better than Condorcet, and when no one is honest, score and range are *phenomenally* better than Condorcet. (You can follow the links from that page for more comprehensive results, but the basic outline is the same for all of them: in every case, at every honest level, score and range outperform all other commonly considered voting systems.)

    Even if you argue–and there is absolutely no data backing this claim up–even if you argue that under your favorite system voter’s will be perfect angels and give up on strategy, while still being devious manipulators under score voting, in some cases you still lose! (As IRV does in the 5-candidate/2-issue example that the graphic is made from.)

    Finally, you’re trying to tell me that Debian and Wikimedia count as “real world”? While the Mathematical Association of America and the American Statistical Association do not? I call bull.

    Range and approval are applicable in any situation where a decision must be made by a group among two or more options–be they judges at ice-skating, movie-aficionados on IMDB, or voters in an election–and statistical simulations have shown that they are the best methods for making such a choice under any conditions varying voter honest, voter knowledge, number of voters, or number of candidates, with a variety of different models for voter decision-space. And you should examine that data before relying on your intuition.

  12. …and you post two more comments while I was responding.

    1. Incumbency is a huge advantage under any electoral system. If you hold approval voting accountable for that while willfully ignoring the exact same effect for all other voting systems, you are being intellectually dishonest.

    2. Mockery will get you nowhere. Allowing voters to vote “for” as many candidates as they would like, rather than for just one as we do today, would not only be simple to implement, it would lead to better electoral outcomes than any more-complicated system you would propose; a result supported by extensive statistical simulation. And again, I’ll take simulation over your misguided intuition.

  13. Well then, while we wait for my far-to-long-and-so-must-await-approval response to come through: Hi.

    I used to live in Pittsburgh as well (moved there for college, stuck around a bit, moved to DC, moved back for a year, and now gone again); I used to larp with Gator and played D&D with Aleris.

  14. I think we’ve both made our respective points and Brady might like to do a post on it or note his observations. I’ve been in Berkeley, California for the past four years, but I still am in touch with folks back in Pittsburgh on Facebook, etc. For what it’s worth I recently had a conversation with one of my Pittsburgh friends about this same subject, not out of academic interest but because California has now replaced party primaries with an open “general” primary from which the top two candidates go on to the November election. Without a ranked ballot strategic voting is going to be ridiculous.

  15. Score Voting and Approval Voting are better than all commonly proposed alternatives. For one thing, they are algorithmically simpler to tabulate than ranked methods (much simpler than even the simplest Condorcet methods). And they are simpler for voters in that ranked methods typically result in something like seven times as many spoiled ballots (like here in my home of San Francisco).

    People have a common misconception that Condorcet methods promote honesty, but tactical exaggeration pays for the individual in most of them.
    http://scorevoting.net/CondBurial.html

    Mike Goldman’s arguments against Approval Voting don’t hold water as far as I can tell. Score and by extension Approval Voting both pass the Favorite Betrayal Criterion, so once you’ve tactically supported your favorite “frontrunner”, you can go ahead and safely support all the candidates you sincerely prefer to that candidate. So say you vote for Gore because you realize he and Bush are the most likely to win — but you really prefer Nader and/or other candidates to Bush AND Gore. Then you can go ahead and support all of them. This is an “optimistic” voting method, because if there’s a candidate that a majority of voters prefer to the frontrunners, he can still win even if voters think he has no chance.

    In most ranked methods, the effective tactic if you prefer Nader>Gore>Bush, is to insincerely vote Gore>Nader>Bush, so these are “cynical”, in that a candidate can’t win if voters don’t think he can win, even if they really prefer him. Explained here for IRV (and same basic logic applies to other ranked systems):
    http://www.electology.org/debate/IrvPlurality

    Finally, Score and Approval Voting are not immune to Arrow’s Theorem and the Gibbard-Satterthwaite theorem, as mathematically proved by a Princeton math Ph.D. named Warren Smith who specializes in voting theory. See the ArrowThm.html page from Dale, as well as this one:
    http://scorevoting.net/GibbSat.html

  16. I meant to say that Score Voting and Approval Voting are not AFFECTED by Arrow’s Theorem or Gibbard-Satterthwaite, or are “immune” to them. I made a typo above and said “not immune”.

  17. “In most ranked methods, the effective tactic if you prefer Nader>Gore>Bush, is to insincerely vote Gore>Nader>Bush, so these are “cynical”, in that a candidate can’t win if voters don’t think he can win”

    I call B.S. The effective vote in this case is Nader>Gore>Bush, because if Nader comes in third place, your vote would be transferred to Gore. Voting Gore>Nader>Bush if you sincerely prefer Nader would be asinine, because it would completely discard any chance of your vote for Nader making any difference.

  18. But that can backfire, and your refusal to vote Gore over Nader could result in Bush, your last choice, winning the election.

    Nine voter/three candidate example for IRV; voters become more “lefty” as we go down the list:

    4: Bush > Gore > Nader
    1: Gore > Bush > Nader
    1: Gore > Nader > Bush
    3: Nader > Gore > Bush <<G>B to G>N>B, then Gore would win (again, run the numbers yourself and see). Which isn’t perfect (you really want Nader!) but is certainly better (you REALLY don’t want Bush).

    The fear of stumbling into precisely this sort of situation is why IRV tends to dominated by the same two parties as under plurality: you’re afraid you’ll cause a spoiler, so you play it safe, and vote for the lesser of two evils. And yes, that’s means your true honest favorite, Nader, will never have any chance to win, just like under plurality.

    Condorcet (and indeed, all ranked-order-ballot based methods) suffer from this or similar problems which tend to drive the electoral system towards two-party domination out of fear of spoilers.

    (The Condorcet example is a bit harder to follow, but I’ll see if I can remember how it goes…)

  19. But that can backfire, and your refusal to vote Gore over Nader could result in Bush, your last choice, winning the election.

    Nine voter/three candidate example for IRV; voters become more “lefty” as we go down the list:

    4: Bush > Gore > Nader
    1: Gore > Bush > Nader
    1: Gore > Nader > Bush
    3: Nader > Gore > Bush *

    Bush wins this election (go ahead, do the runoffs) even though most people prefer Gore over Bush (5:4). But if one of those 3 voters I indicated would instead “cynically lie” and change their vote from N>G>B to G>N>B, then Gore would win (again, run the numbers yourself and see). Which isn’t perfect (you really want Nader!) but is certainly better (you REALLY don’t want Bush).

    The fear of stumbling into precisely this sort of situation is why IRV tends to dominated by the same two parties as under plurality: you’re afraid you’ll cause a spoiler, so you play it safe, and vote for the lesser of two evils. And yes, that’s means your true honest favorite, Nader, will never have any chance to win, just like under plurality.

    Condorcet (and indeed, all ranked-order-ballot based methods) suffer from this or similar problems which tend to drive the electoral system towards two-party domination out of fear of spoilers.

    (The Condorcet example is a bit harder to follow, but I’ll see if I can remember how it goes…)

    (Resumbited: I shouldn’t use less-than so often, silly psuedo-html interpreting thing.)

  20. Condorcet doesn’t suffer from the marginal problem of IRV, which is to say it isn’t a problem at all with ranked ballots but counting methodologies. A ranked ballot is superior to a plurality FPTP, I hope we can all agree on that.

    As for the trivial claim that in a <=3 person contest, approval voting might avoid Gibbard-Satterthwaite and Arrow’s theorem, merely adding a fourth candidate makes that point false.

  21. True; for more than 3 candidates, approval and range do not satisfy G-S.

    But no ranked-order method (including any Condorcet method you know of or could ever conceive of) can satisfy it with more than 2.

    And in case this isn’t clear, 3 is MORE than 2.

    But sure, I will concede that ranked-order-ballot based methods are better than FPTP’s “name one” ballot; the Bayesian regret numbers support that finding. And they also support the finding that approval and score are (under identical conditions for voter distribution, voter honesty, and voter uncertainty) always better than ANY ranked-order-ballot method. Probably because 3 is more than 2.

    Look, I’m not trying to tell you that IRV or Condorcet are strictly WORSE than plurality; I’m trying to tell you that no matter what method you want to use, cardinal ballots are EVEN BETTER than ranked-order ballots. If you want to change electoral systems, let’s do it the BEST way we know how.

  22. @MikeGoldman

    I call B.S. The effective vote in this case is Nader>Gore>Bush, because if Nader comes in third place, your vote would be transferred to Gore. Voting Gore>Nader>Bush if you sincerely prefer Nader would be asinine, because it would completely discard any chance of your vote for Nader making any difference.

    That is a painfully common misconception. See this page for why it is wrong.
    http://www.electology.org/debate/IrvPlurality

  23. Let’s drop the mathematical BS and talk about reality. California has changed its election laws to hold an open primary with the top two finishers going on to the November election. I hope we all agree that with the open primary held as a plurality, we can expect bad results. My proposed reform is to make the open primary a ranked ballot, this may not be a perfect solution, but it is a MODEST solution that has a chance of being passed into law very quickly if a petition were started.

  24. @Mike Goldman,

    Let’s drop the mathematical BS and talk about reality.

    That’s like debating Richard Dawkins about biological evolution and saying, “let’s drop the scientific BS and talk about reality”. Voting is a highly mathematical subject. It’s economics and game theory.

    I hope we all agree that with the open primary held as a plurality, we can expect bad results.

    “Bad” is relative. Score Voting or Approval Voting would be vastly better, but I think top-two runoff is probably better than IRV. TTR has the nice property that in the runoff, voters have no incentive to be insincere, and they get to know the candidates better because there are just two of them. Also the media might as well cover both candidates, since they are the only options — whereas with Plurality Voting, a candidate like Ralph Nader is basically written off since he “can’t win”.

    Consider that Green Party candidate Matt Gonzalez came respectably close to Gavin Newsom in a runoff election for mayor of San Francisco (despite being outspent FIVE-to-ONE), whereas the person who currently holds his old supervisor seat, Ross Mirkarimi, recently switched to the Democrat Party in what some see as a lead up to a mayoral run. There are striking parallels between these two candidates, and yet TTR seems to have been better. That makes sense when you consider how things played out. There was a long stretch of time where Gonzalez was the only alternative to Newsom, so voters had every reason to get to know him. But on our ranked-choice ballots, we typically have many candidates, and even when I get on Youtube and try to learn more about them all, it can be difficult to decide which three to rank.

    I can’t even just rank everyone above Newsom, if I theoretically just want to see some fresh blood, because I only get 3 options. That’s probably largely a result of trying to make a concise and clear IRV ballot. With TTR and Score Voting and Approval Voting, it’s easy to make a concise and clear ballot.

    My proposed reform is to make the open primary a ranked ballot, this may not be a perfect solution, but it is a MODEST solution that has a chance of being passed into law very quickly if a petition were started.

    How are you going to tabulate those ranked ballots? If you use Bucklin voting, that might not be so unreasonable. IRV and/or Condorcet are a bad choice. Approval Voting is simpler than all of them, and certainly far better than IRV.

Leave a Reply to Clay Shentrup Cancel reply

Your email address will not be published. Required fields are marked *

Anti-Spam Quiz: